Inside Ring-1T: Ant engineers solve reinforcement learning bottlenecks at trillion scale

Ant Group, an affiliate of Alibaba, released Ring-1T which it says is the first trillion parameter open-source model.

Nvidia researchers boost LLMs reasoning skills by getting them to 'think' during pre-training

By teaching models to reason during foundational training, the verifier-free method aims to reduce logical errors and boost ...
Hosted on MSN

The Reinforcement Gap — or why some AI skills improve faster than others

This is reinforcement learning (RL), arguably the biggest driver of AI progress ... Some testing kits will work better than others, of course, and some companies will be smarter about how to approach ...
NextBigFuture

Looking at Current AI Learning Frameworks to Create Learning Pipelines to Achieve Superintelligence

Andrej Karpathy says that reinforcement learning is still terrible but better than all other AI learning approaches. Elon ...
Unite.AI

The End of Tabula Rasa: How Pre-Trained World Models are Redefining Reinforcement Learning

For a long time, the core idea in reinforcement learning (RL) was that AI agents should learn every new task from scratch, like a blank slate. This "tabula rasa" approach led to amazing achievements, ...

Mercor quintuples valuation to $10B with $350M Series C

Mercor, which connects AI labs with domain experts for training their foundational AI models, is close to raising $350 million at a $10 billion valuation.
Communications of the ACM

Shields for Safe Reinforcement Learning

Evaluating the advantages and potential drawbacks of shielding as a method for safe RL. Bettina Könighofer is an assistant ...
InfoWorld

3 ways to get into reinforcement learning

Whether you like theoretical study or want to get your hands dirty, plenty of reinforcement learning resources are out there. When I was in graduate school in the 1990s, one of my favorite classes was ...